Upload
vuthu
View
219
Download
0
Embed Size (px)
Citation preview
Behavioural Development Economics:Lessons from Field Labs in theDeveloping World
JUAN CAMILO CARDENAS* & JEFFREY CARPENTER***Facultad de Economıa – CEDE, Universidad de los Andes, Bogota, Colombia, **Department of
Economics, Middlebury College, Vermont, USA
Final version received November 2006
ABSTRACT Explanations of poverty, growth and development depend on the assumptions madeabout individual preferences and the willingness to engage in strategic behaviour. Economicexperiments, especially those conducted in the field, have begun to paint a picture of economicagents in developing communities that is at variance with the traditional portrait. We review thisgrowing literature with an eye towards preference-related experiments conducted in the field. Wealso offer lessons on what development economists might learn from experiments. We conclude bysharing our thoughts on how to conduct experiments in the field and then offer a few ideas forfuture research.
I. Introduction
Experimental economics offers methods to test many behavioural hypotheses at thecore of why, after decades of attempts to induce development with interventionsthrough markets, the state and self-governance, a few countries have escaped povertywhile others remain severely poor. These issues now influence the mainstream ofdevelopment economics (Ray, 1998; Bardhan and Udry, 1999; Hoff and Stiglitz,2001) and behavioural experiments have begun to demonstrate the importance oftesting (and perhaps, replacing) the standard assumptions made about decision-makers. Inspired by our own experiences and those of a growing number ofresearchers, we highlight some of the lessons that can be learned from conductingbehavioural experiments in developing communities. We focus on a topic that hasalways been at the heart of development economics – individual preferences. Welook for insights in four categories of experiments: the propensity to cooperate insocial dilemmas; trust and reciprocity; norms of fairness and altruism; and, risk andtime preferences.
5
10
15
20
25
30
35
40
45
Correspondence Address: Jeffrey Carpenter, Department of Economics, Middlebury College, Vermont,
USA. Email: [email protected] 1
kothandamk 31/1/08 21:14 FJDS_A_284974 (XML)
Journal of Development Studies,Vol. 44, No. 3, 337–364, March 2008
ISSN 0022-0388 Print/1743-9140 Online/08/030337-28 ª 2008 Taylor & Francis
DOI: 10.1080/00220380701848327
Without state support, many developing communities rely on local norms andrules of conduct to provide public goods and regulate the use of common poolresources. Both of these situations can be categorised as social dilemmas in whichindividual incentives are at odds with group incentives. For example, egoisticindividuals could take advantage of the contributions to public goods made byothers (that is, ‘free ride’) but, the community – as a whole – does better wheneveryone contributes. The obvious question is whether the individuals making thesedecisions are egoistic? In addition, the more subtle question is whether egoism isacquired – in other words, do egoism and institutions co-evolve? Our reviewuncovers a considerable amount of cooperation among people in developingcommunities; however, there is variation and although egoists are not the dominanttype, they do exist. One of the most important lessons from this research might bethat cooperative predispositions are heterogeneous and theory needs to account forthis fact.Much of the recent interest in social capital is motivated by the ability of norms of
trust and reciprocity to substitute for formal institutions and complete otherwiseincomplete contracts. Communities rich in trust and reciprocity are thought to bemore productive because the stock of these norms allows people to engage inrelationships that would not be profitable if people were not reciprocally-minded. Aninitial response to this hypothesis was to correlate the performance of countries withsurvey measures of trust and reciprocity (for example Knack and Keefer, 1997).Experiments are now available that allow one to measure these norms moreprecisely. Our survey of the results of such experiments suggests that the variation inexperimental measures of trust and reciprocity are indeed correlated with importanteconomic indicators like the growth rate of GDP, the fraction of the population inpoverty, the rate of unemployment and the Gini coefficient.Developing communities must often adjudicate disputes locally. An interesting
question is whether the norms of fairness and altruism that evolve to fill this voidvary in some systematic way with, for example, the dominant production technology(for example, individual versus team production) and whether this variationcorrelates more generally with the economic performance of the communities. Wereview the distribution experiments conducted in developing communities and,again, find substantial variation in the norms that evolve and the extent to whichthese norms appear to be stable. In many cases the allocations that tend to bepunished as unfair are precisely those allocations that people on the other side of theinteraction tend to avoid.The remaining experiments that we consider, have been constructed to elicit risk
and time preferences. This sort of experiment has a long tradition in developmentliterature (starting with Binswanger, 1980) and has been largely motivated by theproposition that impatience and risk aversion might explain why poor people remainpoor. Our reading of the experimental literature does not support this proposition.We find little evidence that poor people in developing countries are more risk aversethan people in the developed world; however, the impatience results are mixed.Section two is the heart of our review, where we describe the preference
experiments that are now common in the field; and we highlight the important resultsthat might inform the direction of new research on economic development. We aremore critical in section three where we point to a few methodological problems with
50
55
60
65
70
75
80
85
90
338 J. C. Cardenas & J. Carpenter
the current literature. In section three we also offer advice on how to run experimentsin the field. We conclude our review in section four with ideas on how to push thefrontier of preference research in developing communities.
II. Preference Experiments Conducted in Developing Communities
In this section we describe the standard experimental protocols used to elicitpreferences, any universal patterns that we see in the data and, most importantly,lessons of relevance to development economists. Tables 1–5 contain summarystatistics from many of the experiments.
The Propensity to Cooperate in Social Dilemmas
Research on the cooperativeness of individuals in developing countries has usedthree different experiments: the prisoner’s dilemma, the voluntary contributionmechanism, and the common pool resource game. Each game sets up a socialdilemma for the participants in which one strategy leads to the social optimum, whilethe dominant strategy (or best response function) leads to a socially inefficientoutcome. The prisoner’s dilemma (PD) is typically conducted as a symmetric two-person game with two strategies: cooperate and defect, where defect strictlydominates cooperate. PD players are often presented with a normal form matrix inwhich the payoffs are measured in the local currency and, are asked to choosesimultaneously whether to cooperate or defect. However, because interpretingmatrices can pose difficulties for respondents, the game has also been played in theform of a vignette. The voluntary contribution mechanism (VCM) allows players tocontribute to a public good, which has the incentive structure of a n-person prisonersdilemma. This game is repeated for a number of rounds. At the beginning of eachround, participants are given an endowment of tokens that they can place in twoaccounts: a private account that only benefits the decision-maker and a publicaccount that benefits everyone in the group. The amount contributed to the publicaccount is a measure of the cooperativeness of the participant. Lastly, in the commonpool resource game (CPR), players cooperate by not extracting too much from aresource that is accessible to all players but rival or subtract (that is, one player’sextraction reduces the resources available for others to extract). Since this gameusually has the flavour of a non-linear public bad 2, explaining the pay off functiondoes not usually help participants. Instead, they are presented with a table whichrelates their extraction choice to the aggregate choice of the rest of the group.Participants use the table to identify their best response to any extraction taken bythe rest of the group.
Only a minority free ride as a first impulse (Table 1). Approximately one-third ofplayers cooperate in the PD, contributions of half the endowment are common in theVCM and extracting only three-quarters of the Nash level appears to be the norm inthe CPR game. However, there is considerable variation in play. College-agedparticipants in the United States show only moderate rates of cooperation that tendto decline in repeated versions of the VCM, while cooperation rates are higher andsustained among poor participants in Africa and Southeast Asia, suggesting aninverse relationship between norms of cooperation and development; however, the
95
100
105
110
115
120
125
130
135
140
Behavioural Development Economics 339
145
150
155
160
165
170
175
180
185
Table
1.Cooperationin
developingcountries
Game
Study
Location
Students
Meancooperation
PD
Cooper
etal.(1996)
United
States
Yes
22%
cooperate
PD
Hem
esath
andPomponio
(1998)
United
States
Yes
25%
cooperate
China
Yes
54%
cooperate
PD
Tysonet
al.(1988)
South
Africa
Yes
45%
cooperate
w/black
other
Yes
37%
cooperate
w/w
hiteother
VCM
Andreoni(1995)
United
States
Yes
33%
ofendowment
VCM
List(2004)
United
States
No
32%
ofendowment–young
No
43%
ofendowment–old
VCM
Barr
(2001)
Zim
babwe
No
48%
ofendowment,52%
a
VCM
Barr
andKinsey(2002)
Zim
babwe
No
53%
ofendowment–women
Zim
babwe
No
48%
ofendowment–men
VCM
Carpenteret
al.(2004a)
Vietnam
No
72%
ofendowment,76%
a
Thailand
No
61%
ofendowment,73%
a
VCM
Ensm
inger
(2000)
Kenya
No
58%
ofendowment
VCM
Gaechteret
al.(2004)
Russia
Yes
44%
ofendowment
Russia
No
52%
ofendowment
VCM
Henrich
andSmith(2004)
Peru
No
23%
ofendowment
Chile-Mapuche
No
33%
ofendowment
Chile-Huinca
No
58%
ofendowment
VCM
Karlan(2005)
Peru
No
81%
ofendowmentb
CPR
CardenasandCarpenter(2004)
United
States
Yes
79%
ofNash
extraction
Colombia
Yes
74%
ofNash
extraction
CPR
Cardenaset
al.(2000)
Colombia
No
72%
ofNash
extraction
CPR
Cardenas,et
al.(2002)
Colombia
No
68%
ofNash
extraction,49%
c
CPR
Cardenas(2003a)
Colombia
No
74%
ofNash
extraction,62%
c
CPR
Velez
etal.(2006)
Colombia
No
80%
ofNash
extraction
Notes:
a.withoutsocialsanctions,
withsocialsanctions;
b.this
resultis
from
athreshold
publicgoodsgame;
c.withoutcommunication,with
communication.
340 J. C. Cardenas & J. Carpenter
190
195
200
205
210
215
220
225
230
235
Table
2.Trust
indevelopingcountries
Study
Location
Students
Fractionsent
Fractionreturned
Return
ratio
Berget
al.(1995)
United
States
Yes
0.52
0.30
0.90
Burkset
al.(2003)
United
States
Yes
0.65
0.40
1.31
Ashrafet
al.(2005a)
United
States
Yes
0.41
0.23
0.58
Russia
Yes
0.49
0.29
0.80
South
Africa
Yes
0.43
0.27
0.73
Barr
(2003a)
Zim
babwe
No
0.43
0.43
1.28
Buchanet
al.(2003)
United
States
Yes
0.65
0.45a
1.35
China
Yes
0.73
0.50a
1.51
Japan
Yes
0.68
0.50a
1.51
South
Korea
Yes
0.64
0.49a
1.47
Burns(2004b)
South
Africa
Yes
0.33
0.23
0.70
Cardenas(2003b)
Colombia
Yes
0.50
0.41
1.22
Carter
andCastillo
(2002)
South
Africa
No
0.53
0.38
1.14
Castillo
andCarter
(2003)
Honduras
No
0.49
0.42
1.26
Holm
andDanielson(2005)
Tanzania
Yes
0.53
0.37
1.17
Sweden
Yes
0.51
0.35
1.05
DanielsonandHolm
(2003)
Tanzania
No
0.56
0.46
1.40
Ensm
inger
(2000)
Kenya
No
0.44
0.18
0.54
FehrandList(2004)
CostaRica
Yes
0.40
0.32
0.96
CostaRica
No
0.59
0.44
1.32
Greig
andBohnet
(2005)
Kenya
No
0.30
0.41
0.82
Johansson-Stenmanet
al.(2004)
Bangladesh
No
0.46
0.46
1.38
Karlan(2005)
Peru
No
0.46
0.43
1.12
Koford
(2001)
Bulgaria
Yes
0.63
0.46
1.34
Lazzarini,et
al.(2004)
Brazil
Yes
0.56
0.34
0.80
Mosley
andVerschoor(2003)
Uganda
No
0.49
0.33
0.99
Schechter(2004)
Paraguay
No
0.47
0.44
1.31
WilsonandBahry
(2002)
Russia
No
0.51
0.38
1.15
Notes:a.thisfigure
differsfrom
Buchanet
al.(2003)because
they
includethesecond-m
over’sendowmentin
theamountofmoney
availableto
send
back.
Behavioural Development Economics 341
pattern is not perfect. For example, cooperation among poor slash and burnhorticulturalists in Peru is quite low. At the same time, these people and their,relatively, uncooperative neighbours in Chile, do live extremely remotely whichsuggests that, controlling for the frequency of interactions with strangers (seeHenrich, 2000), there may be some relationship between development and theevolution of cooperative norms.1
What other regularities are there? It appears that students are less cooperativethan other participants. However, the difference may be age driven and not really byschooling. Looking only at non-students, List (2004) finds that older people are morecooperative in the United States, as do Gaechter et al. (2004) in Russia andCarpenter et al. (2004b) in southeast Asia. Allowing participants to socially sanction
240
245
250
255
260
265
270
275
280
Table 3. Fairness in developing countries (ultimatum game studies)
Study Location StudentsMean
proposalRejection
rate
Carpenter et al. (2005a) United States Yes 0.41 0.05No 0.45 0.07
Cameron (1999) Indonesia Yes 0.42 0.10Gowdy et al. (2003) Nigeria No 0.43 0.01Henrich et al. (2001) Peru – Machiguenga No 0.26 0.05
Tanzania – Hadza No 0.40, 0.27 0.19, 0.28Bolivia – Tsimane No 0.37 0.00Ecuador – Quichua No 0.27 0.15Mongolia – Torguud No 0.35, 0.36 0.05a
Chile – Mapuche No 0.34 0.07PNG – Au No 0.43, 0.38 0.27, 0.40Tanzania – Sangu No 0.41, 0.42 0.25, 0.05Zimbabwe No 0.41, 0.45 0.10, 0.07Ecuador – Achuar No 0.42 0.00Kenya – Orma No 0.44 0.04Paraguay – Ache No 0.51 0.00Indonesia - Lamelara No 0.58 0.00
Henrich et al. (2006) United States Yes 0.41 0.42b
United States No 0.48 0.71b
Kenya - Maragoli No 0.25 0.96b
Kenya – Samburu No 0.35 0.10b
Kenya – Gusii No 0.40 –Ghana – Accra City No 0.44 0.33b
Tanzania – Hadza No 0.26 0.42b
Tanzania – Isanga No 0.38 0.10b
Siberia – Dolgan No 0.43 0.35b
PNG - Au No 0.44 0.43b
PNG – Sursurunga No 0.51 0.69b
Fiji – Yasawa No 0.40 0.15b
Bolivia – Tsimane No 0.27 0.03b
Colombia – Sanquianga No 0.48 0.30b
Ecuador – Shuar No 0.37 0.10b
Notes: PNG¼Papua New Guinea. Two entries in a cell indicate two different samples in thesame population.a. second rejection rate not reported in the original; b. strategy method used so we report theprobability that the lowest positive offer (10%) would be rejected.
342 J. C. Cardenas & J. Carpenter
free riders also appears to have a noticeable effect on contributions. There is somevariation in the sanctioning technology but in all three cases (Barr, 2001; Carpenter,et al. 2004a), cooperation increases when social sanctions are allowed.2 In the CPRgame, simply allowing participants to discuss the game between rounds has an effectsimilar to social sanctions. For example, Cardenas et al. (2002) show that discussionbetween rounds of the experiment can reduce extraction dramatically, the effect islasting and simple discussion leads to conservation levels that are better than whenimperfect external regulation is imposed. Carpenter et al. (2004b) regress con-tribution levels in a VCM on measures of how often participants have informalconversations with their neighbours and find that more chatty Vietnameseparticipants were also more cooperative.
The Ashraf (2005) experiment complements the monitoring and punishmentstudies. In this experiment, married couples of Filipinos were asked to make familysaving decisions. People were allocated money and have to decide whether to depositit in an account that only the decision-maker can benefit from, or in one that benefitsthe entire family. Disclosure of decision matters were only for the author. However,
285
290
295
300
305
310
315
320
325
Table 4. Altruism in developing countries (dictator game studies)
Study Location Students Mean allocation
Carpenter et al. (2005a) United States Yes 0.25No 0.45
Ashraf et al. (2005a) United States Yes 0.24Russia Yes 0.26South Africa Yes 0.25
Burns (2004a) South Africa Yes 0.26Cardenas and Carpenter (2004) United States Yes 0.27
Colombia Yes 0.19Carter and Castillo (2002) South Africa No 0.42Castillo and Carter (2003) Honduras No 0.42Holm and Danielson (2005) Tanzania Yes 0.24
Sweden Yes 0.28Ensminger (2000) Kenya No 0.31Gowdy et al. (2003) Nigeria No 0.42Henrich et al. (2006) United States Yes 0.32
United States No 0.47Kenya – Maragoli No 0.35Kenya – Samburu No 0.40Kenya – Gusii No 0.33Ghana – Accra City No 0.42Tanzania – Hadza No 0.26Tanzania – Isanga No 0.36Siberia – Dolgan No 0.37PNG - Au No 0.41PNG – Sursurunga No 0.41Fiji – Yasawa No 0.35Bolivia – Tsimane No 0.26Colombia – Sanquianga No 0.44Ecuador – Shuar No 0.35
Note: PNG¼Papua New Guinea.
Behavioural Development Economics 343
330
335
340
345
350
355
360
365
370
375
Table
5.Riskandtimepreferencesin
developingcountries
Decisiontask
Study
Location
Students
Meanbehaviour
Risk:Accept/reject
lotteries
HoltandLaury
(2002)b
USA
Yes
0.685
CRRA
a5
0.97
Risk:Choose
lottery
Binsw
anger
(1980)b
India
No
CRRA
a¼0.71c
Risk:Choose
lottery
Wik
andHolden
(1998)b
Zambia
No
0.815
CRRA
a5
2.0
Risk:Choose
lottery
Barr
(2003b)b
Zim
babwe
No
0.325
CRRA
a5
0.81
Risk:Choose
lottery
Harrisonet
al.(2005)
Ethiopia
No
CRRA
a¼0.05c
India
CRRA
a¼0.84c
Uganda
CRRA
a¼0.17c
Risk:Certainty
equivalent
Barr
andPackard
(2000)
Chile
No
CEd¼0.57
Risk:Certainty
equivalent
Henrich
andMcE
lreath
(2002)
ChileandTanzania
No
CEd(M
apuche)¼0.7
CEd(H
uinca)¼0.4
CEd(Sangu)¼0.68
Risk:Accept/reject
lotteries
Jimenez
(2003)
Spain
Yes
0.405
CRRA
a5
1.25
Risk:Choose
lottery
Nielsen
(2001)
Madagascar
No
CRRA
a¼0.32
Risk:Bet
onadie
roll
Schechter(2005)
Paraguay
No
CRRA
a¼2.57
Tim
e:Accept/reject
delays
Coller
andWilliams(1999)
USA
Yes
17%
5MID
Rf5
20%
Tim
e:Accept/reject
delays
Harrisonet
al.(2002)
Denmark
No
IDR
e¼28%
c
Tim
e:Questionnaire
Barr
andPackard
(2000)
Chile
No
IDR
e¼43%
Tim
e:Accept/reject
delays
Kirbyet
al.(2002)
Bolivia
No
MHID
Rg¼12%
Tim
e:Choose
delay
Nielsen
(2001)
Madagascar
No
IDR
e¼117%
Tim
e:Accept/reject
delays
Pender
(1996)
India
No
MID
R4
50%
Notes:
a.CRRA
isthemeasure
ofconstantrelativerisk
aversion;b.highstakes;c.
controllingfordem
ographics;
d.CE
isthemeancertainty
equivalentconstructed
from
individualfixed
effects
inregressionswitheightobservationsper
participantthatvaried
theoddsofthehighandlow
outcomes.Thismeasure
isexpressed
asafractionofthehighpayout;e.ID
Ristheestimatedindividualdiscountrate;f.Coller
andWilliams(1999)
reportthemedianindividualdiscountrate
(MID
R)because
theiranalysisissensitiveto
cut-offsandthedistributionofresponsesin
right-skew
ed;g.
Kirbyet
al.(2002)report
medianhyperbolicindividualdiscountratesbasedonthefunctionPV¼A/(1þkD)whereA
isthereward,kisthe
hyperbolicdiscountrate,andD
isthedelayin
days.
344 J. C. Cardenas & J. Carpenter
when their wives can find out what they have done, men allocate as much money tothe family as women do but, when they can hide their choices, men are likely to keepmore. These results suggest that models of household bargaining are incomplete andprovide a strong lesson about the use of targeting policy in the Philippines. Cashtransfers and grants, intended to benefit children, for example, should either be givendirectly to mothers or mothers should know when the family has been given money.
A lesson that is starting to emerge from the collection of these cooperationexperiments, is that there seems to be a relationship between the existence of formalinstitutions and the form of norms and social preferences that dictate behaviour.Societies in which formal institutions are missing or weak, develop and rely on pro-social norms and preferences as behavioural benchmarks more than other societies inwhich institutions are strong. This obviously makes sense: I do not need to feel verycooperative towards you if we can write an enforceable contract but when wecannot, then you demonstrating your cooperativeness matters a lot. If informalenforcement mechanisms are provided locally and cooperation is enforced by socialostracism or mutual monitoring, cooperation may be sustainable.
At this stage, the few observations of the interaction between institutions andcooperation suggest more questions than they resolve. For example, while Cardenaset al. (2000) and Carpenter and Seki (2005) provide some evidence that formalinstitutions can crowd out social preferences, the dynamics and, in particular, theprocess of coevolution is not understood very well. Given the possible endogeneity ofnorms and preferences, there may also be a stability-performance trade-off. There isnow plenty of evidence (for example, Ostrom, 1990) suggesting that informalinstitutions may outperform formal institutions because local solutions are oftenbetter informed but, if local norms and rules are less stable than enforced laws thismay not hold.
Other, more concrete, lessons can also be learned from cooperation fieldexperiments. First, the cross-national results on the relationship between age andcooperativeness have implications for fostering collective action, volunteering and,perhaps, fund-raising. Second, the strong results on communication and socialsanctioning should inform the design of institutions. As the data from Barr (2001)suggest, social shaming can have a big effect on behaviour. If one thinks ofmicrocredit as a form of social dilemma, then the success of group lending is not sosurprising given these results.
Group composition is an important predictor of cooperation. For example,Cardenas (2003a) shows how mixing economic classes affects play in a CPR game.Groups composed of mostly poor people conserve common property better thangroups which are mixed between poor people and more affluent local propertyowners. This evidence is contrary to Olson’s (1965) privileged group hypothesis.Likewise, Cardenas and Carpenter (2004) show that mixed groups of students fromdifferent countries in a CPR game perform noticeably worse than homogenousgroups and that these differences are partially explained by conservation attitudes.
At a more basic level, Carpenter et al. (2004a) show how the gender compositionof groups affects the level of cooperation among slum dwelling participants in theVCM and this effect appears to depend on location. In Vietnam, homogeneousgroups of women are more cooperative but, in Thailand it is the men that are morecooperative. If the effects of asymmetries that we document in experiments are
380
385
390
395
400
405
410
415
420
Behavioural Development Economics 345
externally valid, then policies that seem obvious may actually backfire. Extrapolatingfrom Cardenas (2003a), for example, policies that increase the market value oflabour for those people who extract from real CPRs may actually lead to worsemanagement. Instead of relying less on extraction from CPR and, therefore,extracting less, people may just place less value on preserving the resource.
Trust and Reciprocity
In the Berg et al. (1995) investment or trust game (TG), two players are endowedwith money as a show-up fee (typically around $10). The first-mover is given thechance to send as much of her endowment to an anonymous second-mover as shewishes. The experimenter triples the amount of money sent so that sending money issocially efficient. The second-mover then sends back as much he wishes.3
The subgame perfect prediction is straightforward. The second-mover has noincentive to send any money back and therefore, realizing this, the first-mover shouldnot invest anything in the partnership. Despite this prediction, Table 2 shows that inBerg et al. (1995), on average, first-movers send 50 per cent of their endowment andsecond-movers return 30 per cent of what they receive. Despite deviating from theoryin a pro-social direction, sending money is still a bad investment for first-moversbecause they tend to recover only 90 per cent of what they send. In a replication at aninstitution with a significantly smaller and more homogenous student populationhowever, Burks et al. (2003) finds that investing pays off (31% average return).Figure 1 and Table 2 summarise our survey of TG behaviour. The first two letters
in Figure 1 indicate the country in which the experiment was conducted and thename in parentheses indicates the first author of the study. In general, we see that, aswith the cooperation experiments, average play is nowhere near the prediction basedon egoistic preferences. There is an upward sloping relationship between trust andreciprocity, suggesting the possibility of multiple trust-trustworthiness equilibria.
425
430
435
440
445
450
455
460
465
470 Figure 1. An overview of trust game behaviour by country and subject population
346 J. C. Cardenas & J. Carpenter
At one extreme, the South African students in the Ashraf et al. (2005a) study do notsend very much as the first-mover and return significantly less than what is sent tothem as second-movers. At nearly the other extreme, Tanzanian non-students in theDanielson and Holm (2003) study tend to send more than half of their endowmentsand send back a return of 40 per cent, on average.
The norms that communities settle on also differ in efficiency. Greig and Bohnet(2005), for example, show that the underlying norm in the slums of Nairobi isbalanced reciprocity, by which one simply repays an investment as if it were aninterest-free loan. By contrast, most of the data from developed countries supportthe norm of conditional reciprocity in which the two parties see the relationship moreas a partnership in which both players accrue profit. In the first case there is norelationship between trust and reciprocity, as measured by the experiment but, in thesecond, the two are positively correlated. What is important for development,however, is the fact that conditional reciprocity is more socially efficient. Otherexamples of measured community-level differences in the underlying norms drivingbehaviour can be seen in Danielson and Holm (2003) and Carter and Castillo (2003).
Without the Kenyan outlier, the non-students also seem to demonstrate moretrustworthiness. The simple regression of the rate of return on the fraction sent, astudent dummy and the interaction of the two shows that the student intercept issignificantly lower (p5 0.05) and that the student gradient is steeper (p5 0.10). Thissupports what was noted about Russia by Gaechter et al. (2004): student trustexperiments should also be seen as lower bounds on pro-social behaviour.
There are also a few methodological lessons to be learned from these trustexperiments. First, it seems pretty unambiguous that second-mover behaviour in theTG measures trustworthiness or reciprocity, more generally; however, first-moverbehaviour might be harder to interpret. For example, to what extent do first-moverssee their transfer as a donation without any expectation of a return? To examine thispotential altruism, experimenters have followed Cox (2004) and run dictator games,in addition to TGs. In the dictator game (DG), the first-mover simply makes atransfer to a passive second-mover. As there is obviously no self-interested reason totransfer money (the game is anonymous), transfers are interpreted as measures ofaltruism. Assuming that trust and altruism are additively separable, then pure trust isjust the difference between the transfer in the TG and that in the DG. We haveidentified five experiments that conduct both TGs and DGs. Three of them wereconducted in South Africa and find that the amount of ‘pure’ trust is between 21 percent (Carter and Castillo, 2003) and 36 per cent (Ashraf et al. 2005a), with Burns(2004a) finding more pure trust (49%) among high school students. Elsewhere in theworld the estimates also hover around 50 per cent of the amount sent.4
There also now seems to be some consensus that the standard trust question usedin the General Social Survey and the World Values Survey (WVS): ‘Generallyspeaking, would you say that most people can be trusted or that you need to be verycareful in dealing with people?’ is a better measure of trustworthiness than trust.Following Glaeser et al. (2000), Lazzarini et al. (2004) and Johansson-Stenman et al.(2004) regress both trust and trustworthiness on the survey trust response; and, findsthat the survey question correlates much higher with trustworthiness than with trust.In other words, those people who state higher levels of trust in strangers are actuallymore trustworthy. One way to look at this result is as a test of the validity of survey
475
480
485
490
495
500
505
510
515
Behavioural Development Economics 347
trust questions. A related question asks about the validity of the measures elicited inthe TG. The good news is that Karlan (2005), in a Peruvian TG study withparticipants in a group lending programme, does find correlations between gamebehaviour and loan repayment behaviour; however, the results are mixed. On onehand, second-mover ‘trustworthiness’ behaviour appears to be a robust predictor ofrepaying one’s debt but, on the other hand, more ‘trusting’ first-movers are actuallyless likely to repay. Like Schechter (2004), Karlan concludes that first-moverbehaviour might be confounded by risk preferences. Risk-takers are more prone totaking bad risks that leave them unable to repay their loans and these same peopleappear more trusting in the TG.Since the related literature on social capital (see Knack and Keefer, 1997) has
focused on trust and trustworthiness, there has been more interest in running thetrust game in developing communities. This relative wealth of data means we can goa little further in analysing the links between the behavioural measures and economicoutcomes, though such analysis remains tentative.The first step is to examine the link between the WVS measure of trust used by
Knack and Keefer and the experimental measures of trust. As the WVS does notcover six of the countries in the sample, the correlation is based on only 12observations. However, the results are encouraging. The correlation is positive,rho¼ 0.51 and is significant at the 10 per cent level, indicating that countries withmore trust as measured by the WVS, also demonstrate more trust (that is, sendmore) in the TG.The second step is to look for correlations between our experimental measures of
trust and various economic indicators: GDP growth rate, percentage of thepopulation in poverty, Gini coefficients and unemployment rates (Figure 2).5 In eachcase, the correlations are significant at the 5 per cent level or better. Countries withhigher growth rates are associated with more trust (rho¼ 0.51, p¼ 0.02); countrieswith less poverty are associated with more trust (rho¼70.66, p5 0.01); countries inwhich the division of economic gains is more unequal are associated with less trust(rho¼70.48, p¼ 0.04); and, higher unemployment is associated with less trust(rho¼70.64, p5 0.01).
Fairness and Altruism
There are two ways to think about distributional norms that may influence dyadicinteractions. In the simplest case, norms of altruism dictate how generously oneperson must treat another when the second person has little or no power to controlthe outcome. These norms govern many philanthropic acts. Things are a little morecomplicated, however, when the second person has enough power to retaliateagainst perceived injustices. To differentiate the norms that dictate behaviour inthese situations, we use the term, ‘fairness’. Experimenters have developed twosimple games to measure norms of fairness and altruism. One game, the dictatorgame, was defined in the previous section. The second game is the ultimatum game(UG). In the ultimatum game, two players are provisionally allocated a pie to split.The first-mover (proposer) offers a share to the second-mover (responder) whoaccepts or rejects the offer. Accepted offers are implemented; and rejections resultin both players receiving nothing. No responder will choose a rejection threshold
520
525
530
535
540
545
550
555
560
348 J. C. Cardenas & J. Carpenter
larger than zero because she could do better by accepting lower offers. That is,rejecting should be an empty threat. Knowing this, proposers need not offer morethan some small amount. This game has been played hundreds of times indeveloped countries and, while there is some variation in behaviour acrosscountries (see Roth et al., 1991), most behaviour deviates from the subgame perfectequilibrium (of very small offers) in systematic ways. What is interesting from adevelopment perspective is the variety of distributional norms that arise, whetheror not these norms are supported by rejection behaviour and the economic factorsthat determine the norms.
Fairness behaviour in the UG is summarised in Table 3 and, Table 4 summarisesaltruism behaviour in the DG. In each case, the mean allocation to the second personis substantially greater than zero and, in the UG, low offers are routinely rejectedwhich suggests that fairness norms are enforced. If you focus on the subset of casesin which both UG and DG data exist, you can see that while students in the US offer
565
570
575
580
585
590
595
600
605
610
Figure 2. Do experimental measures of trust correlate with economic indicators?
Behavioural Development Economics 349
considerably less when the second person cannot punish, this is not true in general.In Carpenter et al. (2005a) and Henrich et al. (2006), students offer slightly morethan 41 per cent of the pie in the UG, on average but, only, 25 and 32 per cent in theDG, respectively. However, the difference between the mean UG and DG offers inthe developing world tends to be much smaller. This suggests that students in theUnited States are more sensitive to differences in the strategic environment of a gameand the behaviour of other people is more norm-driven. There is another dimensionto the Henrich et al. (2006) data that shows the power of local norms. In some cases,participants in the field reject offers that are too high in addition to offers that are toolow.How do norms and institutions interact? There is little research directly on this
issue. In their study of 15 small scale societies, Henrich et al. (2001) find that twofactors: the local payoffs to cooperation and the degree of market integration explain68 per cent of the variation in UG offers. Those societies in which team work isessential to production (for example, the Lamelara whale fishermen) have strongsharing norms; while societies composed of small bands of isolated and independentfamily groups (for example, the Machiguenga) are not particularly generous towardsoutsiders (nor do they expect the outsiders to be generous). These data also supportHirschman’s (1982) theory of civilising markets. People in societies that are moreintegrated into markets have more experience dealing with strangers and thisexperience seems to foster more fairness.6
Henrich et al. (2006) repeated experiments in a number of the same locations asthe 2001 study. Table 3 shows that the second visit to the Hadza (Tanzania), theTsimane (Bolivia), and the Au (PNG) resulted in data that look very similar to whatwas gathered in 2001. While it is not the same people playing the game, this result is amild version of test-retest reliability and should encourage field experimenters thatthe behaviour measured in these experiments is somewhat robust. There are twoother findings worth mentioning. First, they find that community level differencesexplain much more of the variation in play than individual differences.7 Thissupports the idea that distributional norms are local phenomena and, therefore, mayvary with local economic conditions. Second, these data confirm that distributivenorms are supported by costly punishment. Not only does the rejection behaviour ofsecond-movers in the UG correlate with mean offers by first-movers, using anotherversion of the DG in which an otherwise unaffected bystander can punish thedictator for low offers, the researchers find that the willingness to punish in thisvariant of the DG correlates strongly with offers in the normal version of the DG. Inother words, people are not hypocrites – those who are fairer are more likely to insiston fairness by others.While there is some variation in UG and DG play across communities, in many
places observations with non-student populations pile up on the 50–50 split (forexample, Carpenter et al., 2005a). Although this is interesting in the anthropologicalsense of a human universal, the number of 50–50 splits limits the extent to which theUD and DG can be used to measure fairness norms. To make the games more usefulas instruments, participants need to be ‘pushed off’ the 50–50 split somehow.Recently, Catherine Eckel, Kate Johnson and Duncan Thomas have developed acomparative DG to be used in Mexico. The comparative DG was constructed in thespirit of Sahlins (1972), who proposed a set of concentric social distance circles
615
620
625
630
635
640
645
650
655
350 J. C. Cardenas & J. Carpenter
emanating from the individual. Your family comprises the innermost circle, yourtribe or some other ingroup comprises the next circle and so forth. You extend andexpect different degrees of fairness and reciprocity to and from individuals indifferent circles. In the comparative DG, you are asked to propose an allocation to afamily member, a person in your village and a stranger from a different village.Framing the game in this manner affects behaviour as you would expect. Thesedifferences are more informative than the standard one-decision DG.
The last two lessons that we find in the fairness data are that deservingness is astrong predictor of altruism and social preference experiments seem to also be goodways to measure social interaction effects (Durlauf, 2002). In Fong (2005), studentsplay a DG in which the recipient will be a local welfare recipient. The amount givendepends on whether the potential recipient indicates her willingness to ‘pull herselfup by her bootstraps.’ Those recipients who appear industrious yield significantlymore than those who appear lazy. Similar deservingness effects are found in Branas-Garza (2003) who finds that donations to recipients are approximately six timeslarger when the recipient is identified as poor and in Burns (2004a), in which SouthAfrican participants use school status and race to proxy for deservingness. Castilloand Carter (2003) implement trust and dictator games along with a survey to in-vestigate what Manski (2000) calls social interaction effects. They show how toexploit the variation in behaviour and socio-economic characteristics at both theindividual and group level to identify whether peer effects influence choices.
Time and Risk
A recurring theme in development literature has been that people in underdevelopedcountries are poor partially because they have preferences that are inconsistent withgrowth. They have high discount rates and are risk averse enough, so that it isimpossible for them to save and take the risks necessary to begin to accumulatecapital. For example, Irving Fisher wrote,
A small income, other things being equal, tends to produce a high rate ofimpatience, partly from the thought that provision for the present is necessaryboth for the present itself and for the future as well, and partly from lack offoresight and self-control. (Fisher, 1930: 73)8
In his innovative field study, Binswanger (1980) noted that risk preferencedifferences are important because policy makers can do something about hindrancesto the access of capital but, may be able to do less about the risk attitudes of thosewhom capital would help. In this subsection we consider the evidence on thisconjecture.
Risk experiments fall into two classes which differ in the way that participantsregister their choices. One class is based on the Accept/Reject Lotteries experiment.The most important methodological contribution in this class is Holt and Laury(2002). In this experiment, participants are presented with two columns of pair-wiselottery choices and they must accept one lottery per line and reject the other.Initially, the first column dominates the second in terms of expected payoff andvariance in the payoffs but, eventually, as the probability of the high outcome in the
660
665
670
675
680
685
690
695
700
705
Behavioural Development Economics 351
second column increases, the expected value of the second column starts todominate. Those who are more risk averse will choose the first column for longer.Since this experiment forces participants to choose between two discrete options,their preferences can only be estimated on an interval. On average, Holt and Lauryfind that student participants exhibited levels of constant relative risk aversionbetween 0.68 and 0.97 when they ratchet up the size of the possible payouts which, inthe context of the lotteries offered, is very risk averse behaviour.The second class of risk experiments is the Choose Lottery experiment in which
participants are also presented with a series of lotteries but, in this case, they areasked to pick one from a list which controls for the probability of winning a largeprize (that is, they are all determined by the toss of a coin) but varies the high andlow payouts and, in doing so, the expected payoff. Depending on how risk averse aparticipant is, he should trade off expected return for less variability. Binswanger(1980) was the first to conduct this sort of risk analysis, doing so with peasantfarmers in India. Whilst he does hypothesise that increases in wealth will beassociated with lower risk aversion, this result is not borne out in the regressionanalysis. Despite the differences in the protocols, Binswanger’s average estimate ofconstant relative risk aversion fits within the bounds of the estimate calculated byHolt and Laury. Expanding the comparisons (Holt and Laury, 2002; Binswanger,1980; BarrAT ; Harrison et al.AT ; Jimenez, 2003; Nielsen, 2001; Wik and Holden, 1998)gives us a better idea of whether there are differences between people in developedcountries and those in developing countries. There is some variation in the results(Table 5) but it is not explained by development. In fact, the upper bounds on theHolt and Laury and Jimenez data are larger than the mean values found in thedeveloping world. Overall, there does not appear to be much support for the ideathat poor people in developing countries are more risk averse than richer people indeveloped countries.9
Time preference experiments come in many shapes and sizes. Not only is it difficultto compare studies because of differences in their protocols, there are other issuesthat confound the comparison of time preference data. First, the reported discountrates are very sensitive to how interval choices are interpreted. For example, moststudies use exponential discounting but Kirby et al. (2002) used hyperbolicdiscounting. Further, the number of times that interest is assumed to becompounded per year (obviously) affects the implied discount rate although thisassumption is rarely mentioned. Second, many of these experiments are confoundedby the credibility of the researcher. In most experiments, participants are paid on thespot but by their very nature, time preference experiments must ask people to waitfor their payments. Normally, the researchers are strangers to the participants and,therefore, the experimental data may be biased towards higher discount ratesbecause the participants have two reasons: a preference for the present and nottrusting the experimenter; for choosing a payment today versus a promised paymentin the future. Third, the delays between payments vary with each study which addsone more factor to control for in any analysis.As our developed country benchmarks, consider the estimates of individual
discount rates (IDRs) gathered by Coller and Williams (1999) and Harrison et al.(2002). These experimenters have learned from the past and conduct theirexperiments carefully. For example, to control (at least, partially) for the credibility
710
715
720
725
730
735
740
745
750
352 J. C. Cardenas & J. Carpenter
problem of many experiments, these two papers employ front end delays, whichsimply mean that there is no promise of money today. Instead, people choose, forexample, between money tomorrow and more money in a week. If participants thinkthere is some chance that the experimenters will welch on a promise to pay in thefuture, they should not expect for them to be more likely to welch on a payment oneweek from today than on one that is due tomorrow. Due to the asymmetry of theirdata, Coller and Williams (1999) report median IDRs between 17 and 20 per centusing student data while Harrison et al. (2002), using data from a large sample ofDanes, report mean overall IDRs that control for many demographic factors of 28per cent.
The evidence on whether poor people are more impatient is mixed. Along withtheir risk aversion experiment, Barr and Packard (2000) gather IDRs using ahypothetical questionnaire. Considering the overall variation in estimates, the meanIDR reported in Barr and Packard (43%), is in the ballpark of the developed worldestimates. However, in their fuller analysis, Barr and Packard go on to show thatincome is marginally significantly associated with discount rates (p5 0.10): thosewith larger incomes do appear more patient. Likewise, Kirby et al. (2002), whogather data from the Tsimane horticulturalists of the Bolivian rainforest, show thatdiscount rates are correlated with age (positively), education (negatively) and income(negatively) but, that they are not correlated significantly with wealth. Consideringbehaviours instead of outcomes, Neilsen (2001) AKin a study of peasant farmers onMadagascar, finds a mean IDR of 117 per cent10 and that people who report living inareas in which a lot of deforestation has occurred, have significantly higher discountrates.
Interpretation of the risk and time preference data is a little harder than with thesocial preference data because there are competing models in each domain. Mostpeople, for instance, now believe that the curvature of one’s utility function dependson whether the gamble is formulated in losses or gains. Prospect theory hypo-thesises that people tend to be risk averse in gains but, relatively, risk seeking inlosses (Kahneman and Tversky, 1979). Obviously, because people in developingcountries face real risks and development policies often add to this uncertainty, it isimportant to categorise behavioural predispositions as accurately as possible.Harrison et al. (2005) allow the data to determine the extent to which conventionalexpected utility theory versus prospect theory predicts behaviour and find that thetwo models play roughly equal roles in explaining the behaviour of poor participantsin India, Ethiopia and Uganda. One implication of prospect theory motivatingchoices is that the subjective probability weights that people assign to outcomes maybe very different to the objective weights they should use.
In their estimates, Harrison et al. (2005) find that when participants are told thatan outcome has a 50–50 chance of occurring, they behave as if the chance was as lowas 10 per cent. The authors speculate that recent droughts, in the areas in which theexperiments were conducted, may have accounted for this general pessimism aboutuncertain events. The obvious lesson, however, is that with half the people syste-matically underweighing the likelihood of good outcomes, for instance, it becomeseven harder to rally support for change. The next step is to assess the degree to whichthis probability bias is endogenous and whether simple, first-step, policies would behelpful in altering attitudes towards change.
755
760
765
770
775
780
785
790
795
Behavioural Development Economics 353
There are also two competing models to describe intertemporal decision-making.Economics and finance rely on simple exponential discounting. However, manyrecent studies with students in the developed world have found that models thatassume hyperbolic discounting, work at least as well (see the review of Fredericket al., 2002). Compared to the exponential standard, hyperbolic discountersappear to have inconsistent preferences: they appear impatient in the short termbut extremely patient when considering decisions that will not have to be taken untilsome distant, future date. In other words (that is, those of Elster, 1989), hyperbolicdiscounters suffer from a ‘weakness of will.’ They know what is good for them in thelong run but, when the long run becomes the short run they can not overcome theirimpulses. What are the implications for development policy? It might be the case thatpeople who suffer from weakness of will just need help committing to morefinancially sound decisions. As Mullainathan (2004) suggests, rotating savings andcredit associations (ROSCAs), might serve this purpose. Each member of a ROSCAcontributes an amount to a group fund that goes to one of the members at eachmeeting. Membership requires people to contribute and takes this money out of themember’s hands until it is their turn to receive the fund. While this forced savingspays no interest, it does act as a commitment device that is supported by socialpressure to not free ride.Is there some relationship between poverty, the lack of formal institutions and
hyperbolic discounting? In some sense, this is just a new version of Fisher’sargument – is the commitment problem greater in developing communities thananywhere else? Unfortunately because most economists still assume exponentialdiscounting, there is not a lot of evidence; however, we have found three studies ofrelevance. As a non-student benchmark, Harrison et al. (2002) conclude that, in anationally representative sample of Danes, there is not a lot of evidence to supporthyperbolic discounting. There is heterogeneity but most people have relatively con-stant discount rates over the one to three year horizon. Kirby et al. (2002) estimatehyperbolic discount rates among Amerindian horticulturalists in Bolivia but becausethe maximum delay in their experiment is less than 6 months, the difference betweenthe implied hyperbolic and exponential rates of time preference are too small todetermine which model fits best. By construction, there would be little evidence thatthe Amerindians suffer from weakness of will. In a very cleanly designed fieldexperiment, Pender (1996) returns to the same villages in India in which HansBinswanger conducted his seminal risk experiments. For our purposes, the key resultis that while he does find discount rates to be higher in a 7-month experiment than in a12-month experiment, the 12, 19 and 24-month results are indistinguishable. In otherwords, there is just not enough evidence to make any conclusions about therelationship between development and the prevalence of hyperbolic discounting.In sum, we find little evidence of differences in risk and time preferences between
people in developing and developed economies. Binswanger (1980) suggests thataccess to credit might be much more of a problem than different attitudes to risk.This explanation jibes with other behavioural evidence. For example, neither Barrand Packard (2000) nor Jimenez (2003) find that risk preferences predict whether aperson chooses to be self-employed. Either the risk experiments are not veryexternally valid (which would be contrary to Dohmen et al., 2005) or other factorslike access to credit dominate these decisions.
800
805
810
815
820
825
830
835
840
845
354 J. C. Cardenas & J. Carpenter
III. Running Experiments in the Field
In this section, we reflect on our experiences and on the protocols used by others toformulate a few lessons about running experiments in the field. Other researchershave discussed some of these issues;11 however, we think that our choice of topicsmight be more valuable for development economists. We begin by talking aboutsome of the attributes of the participants and researchers and, then, switch tocomments on the procedural details of running experiments in the field.
There are good reasons why students are known as a ‘convenience’ sample ofthe population. Students are mostly literate, numerate, are somewhat used tothinking abstractly and are accustomed to being told what to do. All theseattributes make the running of decision-making experiments go more smoothly.In the field, however, one should never take literacy or numeracy for granted.Furthermore, experiments can often remind non-student participants of the examsthat caused them to leave school as soon as possible. It is in the interest of theresearcher to make the experiment seem more interesting and straightforwardthan a test.
Simple instructions that rely more on pictures, diagrams and examples are moresuccessful than those using complex grammar and pay off functions. Literacy is alarge problem but it is numeracy that might be something that field experimentersmay stumble on and not be able to recover from. In particular, many people have noidea about the basic laws of probability. For example, Herb Gintis once offered aplausible alternative explanation for why the mean offer in the Henrich (2000) UGrun, with the Machegeunga in Peru, was so low. The published reason has to dowith culture and the low frequency of interacting with strangers. While thisreasoning fits well with the other observations from the 15 small scale societiesproject, a plausible alternative is that nothing is random to people living inAmazonia. Henrich flipped a coin to see if each participant was going to be the first-mover or the second-mover. A coin flip does not mean the same thing to a personwho believes that supernatural forces determine the course of events. Such a personmay think that she was chosen to be the first-mover for a reason and, with thatmindset, demanding most of the pie does not seem unreasonable. Similar reasoningon the part of the person that does not win the toss, may influence one’s willingnessto reject low offers. The point is, however, that these participant attributes all poseadditional problems in the field.
To illustrate the potential importance of numeracy, we can share preliminaryresults from a study of truck driver trainees in the United States. With 749 of thetargeted 1000 observations collected, Stephen Burks, Jeffrey Carpenter, LorenzGoette and Aldo Rustichini found that a standard test of numeracy correlates highlywith trainee behaviour in a variety of experiments. Based on simple correlationcoefficients, more numerate people are less impulsive (p5 0.01), more cooperative asthe first-mover in a sequential PD (p5 0.02), more reciprocal as the second-mover ina sequential PD (p5 0.01), better at backward induction (p5 0.01) and have lowerdiscount rates in the short run (p5 0.01).
We also need to worry about the attributes of the people conducting theexperiments. Both the protocol and the experimenter have to be credible. Fieldexperimenters are rightly criticised by anthropologists for ‘helicoptering’ into a site
850
855
860
865
870
875
880
885
890
Behavioural Development Economics 355
to conduct an experiment, after spending next to no time with, and knowing next tonothing about, the participants. It is reasonable to fear that this behaviour affects thecredibility of the experimenter and may bias the results. A concrete example maybetter illustrate this point. As mentioned in section two, time preference experimentsare often conducted with front-end delays, precisely because the lack of credibility ofthe experimenter will bias the results towards making people appear more impatientthan they really are. Further, if one’s assessment of another’s credibility is, in turn,determined by economic circumstances, the bias could easily be stronger indeveloping countries. One obvious way to mitigate any credibility bias is to spendconsiderable time in the studied communities. This is often not possible but, at aminimum, teaming up with people who have more credibility might be a closesubstitute.Sampling and recruitment is an issue for all of behavioural economics but the
complications arising from non-random sampling and non-representative recruit-ment may be more pronounced in the field. The typical lab recruiting protocol isto run an advertisement in the school newspaper or send out a mass email. Notsurprisingly, this method is not likely to result in a representative sample(Zelenski et al., 2003). For that matter, assignment to treatment is also usuallynot very random. If treatment A happens on Monday and treatment B isscheduled for Tuesday, there could easily be some important unobservable thataffects behaviour and is correlated with the schedules of the students. Theseproblems are likely to be exacerbated in the field because recruitment is ofteneven more chaotic. Given that the researcher has spent a lot of grant money toget to the site, recruitment often becomes like big game hunting. The researchersets up her blind near the ‘watering hole’ and waits for the prey/participants toshow up. As a practical matter, this means that recruitment is much more likelyto happen by word of mouth and, therefore, peer effects, for example, might addto the sampling problems. When friends or relatives show up for a particularsession and the experimental design is highly sensitive to ‘social ties’ (for examplecooperation, trust, public goods, common-pool resource games), one may preferto assign them to different sessions.Given that field participants are likely to find the instructions and protocol
challenging, field experimenters should do what they can to improve how well thegame is understood. Simple things can help increase the quality of the data gatheredin the field. Paper and pen experiments, which are often more difficult to design andtake longer to run, are actually preferred to computerised experiments in the fieldbecause you do not need to worry about computer literacy, on top of the otherparticipant attributes mentioned above. Reading the instructions aloud, usingseveral examples and providing large posters of the decision sheets greatly improvesthe understanding of the participants. We find that examples work well but it is hardto decide which examples to include because examples may also prime participantson one strategy or another. The jury is still out on framing. On the one hand, peopleargue that framing is bad because frames may also cue norm-driven behaviour andwe are looking for robust behaviours that may not be situation-specific. On the otherhand, mild frames may assure that all the participants are playing the game that theexperimenter intended. In other words, it may be more important for everyone to be‘on the same page.’
895
900
905
910
915
920
925
930
935
940
356 J. C. Cardenas & J. Carpenter
Henrich et al. (2006) have used the strategy method in which participants are askedto provide a full strategy of behaviour instead of responding to whatever stimulusthey encounter. A second-mover, for example, in the UG provides a full strategy bystating which offers she will accept and which offers she will reject, rather thanaccepting or rejecting the offer that she is presented with. The benefit is obvious; theexperimenter receives much more information. However, the potential problem withthis method of eliciting responses is that many people do not naturally think in termsof strategies and, therefore, it is hard for them to think of what they would do inevery possible situation, while it is easy to think about what they will do in thesituation in which they find themselves.
In some settings, cash payments are less desirable than payment in durables. Indeveloping communities that are not very well integrated into markets, cash is notparticularly appealing while tools and other durables are. Hence, to properlyincentivise participants, one should be sensitive to their needs. For example, in astudy of the rural poor in Andhra Pradesh, Pender (1996) used rice as paymentinstead of cash. Regardless of the incentive type, paying on average one to two dayswage for a half-day session seems to create the necessary salience for participants inthe field.
A final concern is cross talk; that is when one set of participants talks about theexperiment to another set that have not yet participated. This can be a problembecause experiments that happen over a sequence of days may implicitly set up asystem of overlapping generations so that later choices are biased by earlierbehaviour. Cross talk may also introduce selection problems because it could affectwho participates. Preventing cross talk is a major challenge. Some people try torecruit large numbers of participants to run concurrent sessions and, other peoplebuild in waiting periods, to separate the people who already have participatedfrom those that have not. In either case, experimenters should keep record of theexact day and time of each session in order to keep track of possible cross talkeffects.
IV. The Frontiers of Preference Research in LDCs
We conclude by offering a few thoughts on the future of preference-relatedexperimental research in developing countries. We begin by discussing each of theareas covered in section two before ending with a discussion of a few other relatedideas.
We still have limited knowledge of the determinants, stability or implications ofsocial preferences. However, the little that we do know suggests that economicoutcomes might be sensitive to the distribution of social preferences. Considering thepropensity to cooperate in social dilemmas, we have just begun to understand therelationship between formal institutions (laws, in particular) and informal norms ofcooperation. We should know more about the co-evolution of formal and informalrules to promote group welfare.
The experiments we have surveyed suggest that information is an importantdeterminant of individual cooperativeness because most people are, to some degree,conditionally cooperative. Information cues the norm of conditional cooperationdirectly and affects the expectations that allow simultaneous games to be solved
945
950
955
960
965
970
975
980
985
Behavioural Development Economics 357
cooperatively. We should conduct research to better understand the information thathelps promote and sustain the social preferences that support cooperative behaviourand collective action.Once the basic trust game has been run in more countries, it will be possible to
attempt replication of the work of Knack and Keefer (1997), Zak and Knack (2001)using behaviourally validated data. Focusing at the community level will also alloweconomists to test the hypothesised links between trust, trustworthiness and theexistence of poverty traps. The poverty trap explanation of low growth is slightlydifferent than the standard linear social capital model. Poverty traps are thought toexist because there are multiple stable trust-growth equilibria. If these models betterdescribe reality, then the key is to shock the system out of the basin of attraction ofthe low trust–low growth equilibrium.Rosenzweig (1988) concludes that extended families are often substitute for formal
institutions in developing countries. Disputes are more likely to be adjudicatedthrough the family than through a formal system of courts. Bonds of commonexperience and altruism enable families to transcend the informational barriers thattend to hinder the development of formal institutions. If this is indeed the case, thenfield experiments should be the obvious way to identify the norms of fairness andaltruism at the core of these family structures. Once the norms have been identified,it will be possible to examine if variation in norms accounts for important economicindicators like education attainment and income.The evidence that we have gathered suggests that poor people are not more risk
averse (or risk-loving) than rich people and there is mixed evidence on whether poorpeople are more impatient than rich people. More data would allow us to layarguments, such as those of Fisher to rest, once and for all. A standardised protocolshould be developed based on the latest research from the field and experimentsshould be conducted in ‘sister cities’ straddling the development divide that controlfor many of the differences (for example, distance to a major city and populationdensity) that confound current comparisons.Mullainathan (2004) lists a number of development puzzles (for example, poverty
traps, non-optimal resource use) that might be explained by hyperbolic discounting.However, the data does not currently exist to test whether poor people are morelikely to discount the present more than the future. Not only do we not know theextent to which people in developing communities discount hyperbolically, we alsodo not have much knowledge about the degree to which this preference isendogenous. In what sense might the conditions of poverty cause hyperbolic dis-counting? What is the impact of education on the propensity to discounthyperbolically? In the spirit of Ashraf et al. (2005b), what policies and contractscan be offered to help poor people commit to savings, given they may discounthyperbolically?Finally, in many places, we have pointed to the interaction and tension between
formal and informal institutions. We consider this to be another area in which thereturn to research will be high. People traditionally think of informal institutions asexisting before formal institutions are established; however, we hope that this is nolonger obvious to the reader. We think that there are plenty of cases in which formaland informal institutions coexist; there are cases in which formal institutions crowdout informal institutions; and, there are even cases in which informal institutions fill
990
995
1000
1005
1010
1015
1020
1025
1030
358 J. C. Cardenas & J. Carpenter
the void following the decline of formal institutions. Experiments can help furtheranalysis of the relationship between formal and informal institutions. Is it actuallythe case (as proposed by Hirschman, 1982), for example, that markets – as formalinstitutions – simply replace other informal economic relationships?
Our sense is that most development economists are still wary of the use ofexperiments. While being strong advocates, we understand the reticence –behavioural economists are only beginning to give other practitioners a reason tocare. This issue is often linked to the idea of external validity: do the behaviouralpropensities that we capture in experiments correlate with economic activity outsidethe field lab? Other economists may also be sceptical of the process of recruitmentand sampling. A major methodological step will be to work to make our samplesmore representative and, to reduce the self-selection bias that can arise fromvoluntary participation. Making sure that the incentives used correspond to theopportunity cost of participating and, at a minimum, collecting demographics frompeople who decline participation should help.
As the reader can see, there is already a substantial body of experimental work onpreferences in developing countries. While some questions have been answered, themajor contribution of the current literature is the establishment of a fieldmethodology. We also think that these methods will now set the stage for new,more policy-oriented research. If policies are aimed at inducing changes in behaviourto improve outcomes, experiments can provide detailed behavioural data about theeffects of certain incentives, institutions or information on a specific context orgroup. In the future, these data may also make it feasible to calibrate and tailorpolicies at the local level.
We conclude by summarising some policy relevant findings. Experimental work onthe distribution of resources within a household is extremely important, given thatgrantors typically assume conditional cash transfers must be put in the hands ofmothers, to get resources to children. An example is the Progressa programme inMexico. Field experiments will test the assumptions at the foundation of thisprogramme. Must grantors target women because they are more altruistic towardstheir children or, is it because they are more likely to invest in the future of theirchildren?
How might the heterogeneity of distributional norms affect policy? As ElinorOstrom pointed out, some policies function on the basis of state or other third partyintervention and others arise endogenously (Ostrom, 1998). As her work suggests,the right policy often depends on community attributes like the strength andcomposition of local norms. In this sense, local distributive customs often correlatewith the extent to which intervention helps the situation or makes it worse.Moreover, the relationship between the nature of institutions (formal versusinformal) and the potency of an intervention, extends to situations in which trust andreciprocity (section two) support more efficient outcomes.
Experiments on time and risk preferences will also continue to influence policy.Measuring these preferences allows policy-makers to first assess the extent to whichpreferences hinder accumulation. So far, there is little evidence that people remainpoor because they are risk averse and the evidence on the relationship with timepreferences is mixed. However, once the preferences have been measured moreprecisely, the results may point to more effective policies. If it is the case that
1035
1040
1045
1050
1055
1060
1065
1070
1075
1080
Behavioural Development Economics 359
preferences act as constraints, then commitment mechanisms like ROSCAs, in thecase of impatience, should be explored; and for constraints due to attitudes towardsrisk, farmers should be educated in the actual riskiness of new technologies, ratherthan to continue to allow the ambiguity of choice to paralyse them (Engle-Warnicket al., 2006). On the other hand, if preferences do not correlate with poverty, it istime to redouble efforts on providing access to credit, for example.
Acknowledgements
We thank Hans Binswanger, Sam Bowles, Colin Camerer, Karla Hoff, DuncanThomas, the editor, Howard White and two anonymous referees for thoughtfulcomments on previous drafts. We also acknowledge the financial support of aResearch and Writing grant from the MacArthur Foundation, the MacArthurFoundation’s Norms and Preferences Working Group and the National ScienceFoundation (SES-CAREER 0092953). Cardenas thanks the International Fellow-ship programme at the Santa Fe Institute and Natalia Candelo for keeping the tablesup to date.
Notes
1. Of course all this speculation is intended to be provocative and needs to be weighed against other
explanations for the variation in play. One that is obvious and should be remembered when comparing
many of the results we review, is that protocols are not standardised in these experiments and the
details of the game do often matter.
2. Barr (2001) allowed participants to directly voice disapproval while participants in Carpenter et al.
(2004a) paid a cost to send a message (a picture of an unhappy face) to the rest of the group, when they
were unhappy with contributions.
3. This is how Berg et al. (1995) ran the game but other variations have been seen. For example, Glaeser
et al. (2000) do not give an endowment to the second-mover and double, rather than triple, the
transfer.
4. First-movers may also view the game as more of a gamble than an exchange. The potential risk
confound has been discussed in Karlan (2005) and examined in Schechter (2004). Of course, such a
correlation is not too surprising but it does suggest that insuring against risk might be as effective as
building trust.
5. There are fewer observations in the unemployment graph because this information is not reported for
Tanzania or Uganda.
6. Of course the sort of farmers’ markets that the Henrich et al. (2001) participants are used to, are vastly
different than the thick anonymous markets common in developed economies. It is not obvious that
the effect of market integration is globally monotonic, especially controlling for the size of the market.
7. Gowdy et al. (2003) and Bahry and Wilson (2003) draw similar conclusions.
8. More recently, this fable has been discussed in Lawrance (1991), Moseley (2001), Neilsen (2001)AK and
Ogaki and Atkeson (1977).
9. We point out but do not discuss, two other studies that do not fit neatly into our two categories of risk
experiments. Barr and Packard (2000) adopt the Schubert et al. (1999) method for measuring the
certainty equivalents of Chilean adults to determine if the self-employed are less risk averse. Schechter
(2005) had people in rural Paraguay bet on the roll of a die to test their attitudes towards risk. Perhaps
because the protocol is sufficiently different, Schechter finds rather high levels of risk aversion (the
average CRRA was 2.57).
10. However, there appear to be a number of issues with Neilsen’s experiment.AK One problem is that the
lowest IDR interval is between 0 and 20 per cent. Another problem is that the intervals are so wide
that little precision can be expected.
11. See Roth et al. (1991), Harrison and List (2004) and Carpenter et al. (2005b).AL
1085
1090
1095
1100
1105
1110
1115
1120
1125
360 J. C. Cardenas & J. Carpenter
References
Andreoni, J. (1995) Warm-glow versus cold-prickle: the effects of positive and negative framing on
cooperation in experiments. Quarterly Journal of Economics, 110(1), pp. 1–21.
Ashraf, N. (2005) Spousal control and intra-household decision making: an experimental study in the
Philippines. Working paper, Harvard Business School.
Ashraf, N., Bohnet, I. and Piankov, N. (2005a) Decomposing trust. Experimental Economics,
forthcoming. 3Ashraf, N., Karlan, D. and Yin, W. (2005b) Tying Odysseus to the mast: evidence from a commitment
savings product in the Philippines. Quarterly Journal of Economics, forthcoming. 3Bahry, D. and Wilson, R. (2003) Confusion or fairness in the field? Rejections in the ultimatum game
under the strategy method. Working paper, Department of Political Science, Rice University.
Bardhan, P. and Udry, C. (1999) Development Microeconomics (Oxford and New York: Oxford University
Press).
Barr, A. (2001) Social dilemmas and shame-based sanctions: experimental results from rural Zimbabwe.
Working paper, Centre for the Study of African Economies. 4Barr, A. (2003a) Trust and expected trustworthiness: experimental evidence from Zimbabwe. The
Economic Journal, 113(489), pp. 614–630.
Barr, A. (2003b) Risk pooling, commitment, and information: an experimental test of two fundamental
assumptions. Working Paper CSAE WPS/2003–05, Centre for the Study of African Economies,
Oxford University.
Barr, A. and Packard, T. (2000) Revealed and concealed preferences in the Chilean pension system:
an experimental investigation. Discussion Paper Series, Department of Economics, University of
Oxford.
Barr, A. and Kinsey, B. (2002) Do men really have no shame? Working paper, Centre for the Study of
African Economies. 4Berg, J., Dickaut, J. and McCabe, K. (1995) Trust, reciprocity and social history. Games and Economic
Behavior, 10, pp. 122–142. 5Binswanger, H. (1980) Attitudes toward risk: experimental measurement in rural India. American Journal
of Agricultural Economics, 62(August), pp. 395–407.
Branas-Garza, P. (2003) Poverty in dictator games: awakening solidarity. Working paper E2003/50,
centrA. 6Buchan, N., Croson, R. and Johnson, E. (2003) Let’s get personal: an international examination of the
influence of communication. Culture, and social distance on other regarding preferences. Journal of
Economic Behavior and Organization, 51. 3Burks, S.V., Carpenter, J.P. and Verhoogen, E. (2003) Playing both roles in the trust game. Journal of
Economic Behavior and Organization, 51(2), pp. 195–216.
Burns, J. (2004a) Inequality aversion and group identity in segmented society: evidence from South Africa.
Working paper, Department of Economics, University of Cape Town.
Burns, J. (2004b) Rebuilding trust in post-apartheid South Africa. Working paper, Department of
Economics, University of Cape Town.
Cameron, L. (1999) Raising the stakes in the ultimatum game: experimental evidence from Indonesia.
Economic Inquiry, 37(1), pp. 47–59.
Cardenas, J.C. (2003a) Real wealth and experimental cooperation: evidence from field experiments.
Journal of Development Economics, 70, pp. 263–289. 5Cardenas, J.C. (2003b) En vos cofio: an experimental exploration on the micro-foundations of trust,
reciprocity, and social distance in Colombia. Working paper, Faculty of Economics, Universidad de
Los Andes.
Cardenas, J.C. and Carpenter, J. (2004) An inter-cultural examination of cooperation in the commons.
Working paper, Department of Economics, Middlebury College.
Cardenas, J.C., Stranlund, J. and Willis, C. (2000) Local environmental control and institutional
crowding-out. World Development, 28(10), pp. 1719–1733.
Cardenas, J.-C., Stranlund, J. and Willis, C. (2002) Economic inequality and burden-sharing in the
provision of local environmental quality. Ecological Economics, 1–17. 7Carpenter, J. and Seki, E. (2005) Do social preferences increase productivity? Field experimental evidence
from fishermen in Toyama bay. IZA discussion paper 1697.
1130
1135
1140
1145
1150
1155
1160
1165
1170
1175
Behavioural Development Economics 361
Carpenter, J., Daniere, A. and Takahashi, L. (2004a) Social capital and trust in southeast Asian cities.
Urban Studies, 41(4), pp. 853–874.
Carpenter, J., Daniere, A. and Takahashi, L. (2004b) Cooperation, trust, and social capital in southeast
Asian urban slums. Journal of Economic Behavior and Organization, 55(4), pp. 533–551.
Carpenter, J., Burks, S. and Verhoogen, E. (2005a) Comparing students to workers: the effects of social
framing on behavior in distribution games, in: J. Carpenter, G. Harrison and J. List (eds) Field
Experiments in Economics, (Greenwich, CT and London: JAI/Elsevier), pp. 261–290.
Carter, M. and Castillo, M. (2002) The economic impacts of altruism, trust and reciprocity: an
experimental approach to social capital. AAE Staff Papers, University of Wisconsin-Madison.
Carter, M. and Castillo, M. (2003) An experimental approach to social capital in South Africa. Working
paper, Department of Agricultural Economics, University of Wisconsin.
Castillo, M. and Carter, M. (2003) Identifying social effects in economic field experiments. Working paper,
Department of Agricultural and Applied Economics, University of Wisconsin.
Coller, M. and Williams, M.B. (1999) Eliciting individual discount rates. Experimental Economics, 2(2),
pp. 107–127.
Cooper, R., DeJong, D., Forsythe, R. and Ross, T. (1996) Cooperation without reputation: experimental
evidence from prisoner’s dilemma games. Games and Economic Behavior, 12, pp. 187–218.5Cox, J.C. (2004) How to identify trust and reciprocity. Games and Economic Behavior, 46, pp. 260–281.5Danielson, A. and Holm, H. (2003) Do you trust your brethren? Eliciting trust attitudes and trust behavior
in a Tanzanian congregation, ????.8Dohmen, T., Falk, A., Huffman, D., Sunde, U., Schupp, J. and Wagner, G. (2005) Individual risk
attitudes: new evidence from a large, representative, experimentally-validated survey. IZA discussion
paper 1730.
Durlauf, S.N. (2002) On the empirics of social capital. The Economic Journal, 112(483), pp. 459–479.
Elster, J. (1989) The Cement of Society: A Study of Social Order, (Cambridge: Cambridge University
Press).
Engle-Warnick, J., Escobal, J. and Laszlo, S. (2006) Risk preference, ambiguity aversion and technology
choice: experimental and survey evidence from Peru, ????.8Ensminger, J. (2000) Experimental economics in the bush: why institutions matter, in: C. Menard (ed.)
Institutions, Contracts, and Organizations: Perspectives from New Institutional Economics, (London:
Edward Elgar), pp. 158–171.
Fehr, E. and List, J. (2004) The hidden costs and returns of incentives: trust and trustworthiness among
CEOs. Journal of the European Economic Association, 2(5), pp. 743–771.
Fisher, I. (1930) The Theory of Interest, (New York: Macmillan).
Fong, C. (2005) Empathic responsiveness: evidence from a randomized experiment on giving to welfare
recipients. Working paper, Department of Social and Decision Sciences, Carnegie Mellon.
Frederick, S., Loewenstein, G. and O’Donoghue, T. (2002) Time discounting and time preference: a
critical review. Journal of Economic Literature, 40(2), pp. 351–401.
Gaechter, S., Herrmann, B. and Thoeni, C. (2004) Trust, voluntary cooperation, and socio-economic
background: survey and experimental evidence. Journal of Economic Behavior and Organization, 55(4),
pp. 505–531.
Glaeser, E., Laibson, D., Scheinkman, J. and Soutter, C. (2000) Measuring trust. The Quarterly Journal of
Economics, 65(3), pp. 811–846.
Gowdy, J., Iorgulescu, R. and Onyeiwu, S. (2003) Fairness and retaliation in a rural Nigerian village.
Journal of Economic Behavior and Organization, 52, pp. 469–479.5Greig, F. and Bohnet, I. (2005) Is there reciprocity in a reciprocal-exchange economy? Evidence from a
slum in Nairobi, Kenya. Working paper, Kennedy School of Government.
Harrison, G. and List, J. (2004) Field experiments. Journal of Economic Literature, 42(December),
pp. 1009–1055.
Harrison, G., Humphrey, S. and Verschoor, A. (2005) Choice under uncertainty in developing countries.
CeDEx Discussion paper no. 2005–18.
Harrison, G.W., Lau, M.I. and Williams, M.B. (2002) Estimating individual discount rates in Denmark: a
field experiment, American Economic Review, 92(5), pp. 1606–1617.
Hemesath, M. and Pomponio, X. (1998) Cooperation and culture: students from China and the United
States in a prisoner’s dilemma, Cross-Cultural Research: The Journal of Comparative Social Science,
32(2), pp. 171–184.
1180
1185
1190
1195
1200
1205
1210
1215
1220
362 J. C. Cardenas & J. Carpenter
Henrich, J. (2000) Does culture matter in economic behavior? Ultimatum game bargaining among the
Machiguenga indians of the Peruvian Amazon, American Economic Review, 90(Sept), pp. 973–979.
Henrich, J. and McElreath, R. (2002) Are peasants risk-averse decision makers? Current Anthropology,
43(1), pp. 172–181.
Henrich, J. and Smith, N. (2004) Comparative experimental evidence from Machiguenga, Mapuche,
Huinca and American populations shows substantial variation among social groups in bargaining and
public goods behavior, in: J. Henrich, R. Boyd, S. Bowles, H. Gintis, E. Fehr and C. Camerer (eds)
Foundations of Human Sociality: Economic Experiments and Ethnographic Evidence from fifteen Small-
scale Societies, (Oxford: Oxford University Press), pp. 125–167.
Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H. and McElreath, R. (2001) In search of
homo economics: behavioral experiments in 15 small-scale societies, American Economic Review,
91(2), pp. 73–78.
Henrich, J., McElreath, R., Barr, A., Ensminger, J., Barrett, C., Bolyanatz, A., Cardenas, J.C., Gurven,
M., Gwako, E., Henrich, N., Lesorogol, C., Marlowe, F., Tracer, D. and Ziker, J. (2006) Costly
punishment across human societies, Science, 312(23 June), pp. 1767–1770.
Hirschman, A. (1982) Rival interpretations of market society: civilizing, destructive, or feeble? Journal of
Economic Literature, 20(Dec), pp. 1463–1484.
Hoff, K. and Stiglitz, J. (2001) Modern economic theory and development, in: G. Meier and J. Stiglitz
(eds) Frontiers of Development Economics, (Oxford: Oxford University Press), pp. 389–459.
Holm, H. and Danielson, A. (2005) Trust in the tropics? Experimental evidence from Tanzania, The
Economic Journal, 115(503), pp. 505–532.
Holt, C. and Laury, S. (2002) Risk aversion and incentive effects, American Economic Review, 92(5),
pp. 1644–1655.
Jimenez, F. (2003) Does familiar environment affect individual risk attitudes? Olive-oil producer vs. no-
producer households. Documento de trabajo Serie Economia E2003/28, centrA. 6Johansson-Stenman, O., Mahmud, M. and Martinsson, P. (2004) Trust, trust games and stated trust:
evidence from rural Bangladesh. Working paper, Department of Economics, Goteborg University.
Kahneman, D. and Tversky, A. (1979) Prospect theory: an analysis of choice under risk. Econometrica, 47,
pp. 263–291. 5Karlan, D. (2005) Using experimental economics to measure social capital and predict financial decisions,
American Economic Review, 95(5), pp. 1688–1699.
Kirby, K., Godoy, R., Reyes-Garcia, V., Byron, E., Apaza, L., Leonard, W., Perez, E., Vadez, V. and
Wilkie, D. (2002) Correlates of delay-discount rates: evidence from Tsimane amerindians of the
Bolivian rain forest, Journal of Economic Psychology, 23, pp. 291–316. 5Knack, S. and Keefer, P. (1997) Does social capital have an economic payoff? A cross-country
investigation, Quarterly Journal of Economics, 112, pp. 1251–1288. 5Koford, K. (2001) Trust and reciprocity in Bulgaria: a replication of Berg, Dickhaut and McCabe (1995).
Working paper, University of Delaware.
Lawrance, E. (1991) Poverty and the rate of time preference: evidence from panel data, Journal of Political
Economy, 99(1), pp. 54–77.
Lazzarini, S., Madalozzo, R., Artes, R. and de Oliveira Siqueira, J. (2004) Measuring trust: an experiment
in Brazil. Ibmec working paper – WPE – 2004 – 1.
List, J. (2004) Young, selfish and male: field evidence of social preferences, The Economic Journal,
114(Jan), pp. 121–149.
Manski, C. (2000) Economic analysis of social interactions, Journal of Economic Perspectives, 14(3),
pp. 115–136.
Moseley, W. (2001) African evidence on the relation of poverty, time preference and the environment,
Ecological Economics, 38, pp. 317–326. 5Mosley, P. and Verschoor, A. (2003) The development of trust and social capital in rural Uganda: an
experimental approach. Working paper, University of Sheffield.
Mullainathan, S. (2004) Psychology and development economics. Working paper, Department of
Economics, MIT.
Nielsen, U. (2001) Poverty and attitudes towards time and risk - experimental evidence from Madagascar.
Working paper, Royal Veterinary and Agricultural University of Denmark.
Ogaki, M. and Atkeson, A. (1977) Rate of time preference, intertemporal elasticity of substitution, and
level of wealth, Review of Economics and Statistics, 79(4), pp. 564–572.
1225
1230
1235
1240
1245
1250
1255
1260
1265
Behavioural Development Economics 363
Olson, M. (1965) The Logic of Collective Action, (Cambridge: Harvard University Press).
Ostrom, E. (1990) Governing the Commons: The Evolution of Institutions for Collective Action, (Cambridge:
Cambridge University Press).
Ostrom, E. (1998) A behavioral approach to the rational choice theory of collective action presidential
address. American Political Science Review, 92(1), pp. 1–22.
Pender, J.L. (1996) Discount rates and credit markets: theory and evidence from rural India, Journal of
Development Economics, 50(2), pp. 257–296.
Ray, D. (1998) Development Economics, (Princeton: Princeton University Press).
Rosenzweig, M. (1988) Risk, implicit contracts and the family in rural areas of low-income countries. The
Economic Journal, 98(Dec), pp. 1148–1170.
Roth, A., Prasnikar, V., Okuno-Fujiwara, M. and Zamir, S. (1991) Bargaining and market behavior in
Jerusalem, Ljubljana, Pittsburgh and Tokyo: an experimental study, American Economic Review, 81(5)
(Dec), pp. 1068–1095.
Sahlins, M. (1972) Stone Age Economics, (Chicago: Aldine).
Schechter, L. (2004) Traditional trust measurement and the risk confound: an experiment in rural
Paraguay, Journal of Economic Behavior and Organization, forthcoming.3Schechter, L. (2005) Risk aversion and expected-utility theory: a calibration exercise. Working paper,
Department of Agricultural Economics, University of Wisconsin.
Schubert, R. and et al. (1999) Financial decision-making: are women really more risk-averse? American
Economic Review, 89(2), pp. 381–385.9Tyson, G.A., Schlachter, A. and Cooper, S. (1988) Game playing strategy as an indicator of racial
prejudice among South African students, Journal of Social Psychology, 128(4), pp. 473–485.
Velez, M.A., Stranlund, J. and Murphy, J. (2006) What motivates common pool resource users?
Experimental evidence from the field. Working paper, Department of Resource Economics, University
of Massachusetts.
Wik, M. and Holden, S. (1998) Experimental studies of peasant’s attitudes towards risk in northern
Zambia. Working Paper D-14, Agricultural University of Norway.
Wilson, R. and Bahry, D. (2002) Intergenerational trust in transitional societies. Working paper,
Department of Political Science, Rice University.
Zak, P.J. and Knack, S. (2001) Trust and growth, Economic Journal, 111(470), pp. 295–321.
Zelenski, J., Rusting, C. and Larsen, R. (2003) Consistency in the time of experiment participation and
personality correlates: a methodological note, Personality and Individual Differences, 34(4), pp. 547–
558.
1270
1275
1280
1285
1290
1295
1300
1305
1310
1315
364 J. C. Cardenas & J. Carpenter